Tiling array data analysis: a multiscale approach using wavelets.

Area of Science:

• Genomics
• Signal Processing
• Bioinformatics

Background:

• Tiling array data from ChIP-chip experiments is often obscured by noise, hindering accurate interpretation.
• Wavelet transformation is a powerful signal processing tool for isolating true signals from noisy datasets.
• Coiflet wavelets were chosen for their shape, closely matching expected ChIP-chip peak profiles.

Purpose of the Study:

• To denoise representative ChIP-chip datasets using wavelet transformation.
• To develop a robust thresholding procedure for ChIP-chip data analysis.
• To improve the interpretability and accuracy of ChIP-chip experimental results.

Main Methods:

• Application of wavelet transformation, specifically Coiflet basis functions, to ChIP-chip data.
• Analysis of wavelet coefficients to differentiate noise from biological signals.
• Development of a thresholding procedure based on the log-normal distribution of noise coefficients.

Main Results:

• Noise was found to be concentrated at small scales, while true signals spanned larger scales in wavelet-transformed data.
• Wavelet coefficients from non-specific cross-hybridization followed a log-normal distribution, enabling precise thresholding.
• The developed algorithm successfully applied to diverse ChIP-chip datasets, including those with broad peaks.

Conclusions:

• The wavelet-based method allows for unambiguous, absolute threshold setting in ChIP-chip experiments.
• Benchmarking against other methods using spike-ins on the ENCODE Nimblegen tiling array showed superior performance.
• Wavelet denoising achieved the best overall score, demonstrating its effectiveness for ChIP-chip data analysis.